CCSS: Quickest Detection Under Energy Constraints

无线传感器网络通常用于监测周围环境的异常变化。这些变化通常意味着某些严重后果的活动，如结构故障或化学品/气体泄漏等。最快检测是一个框架，专注于设计顺序检测算法，以尽可能快速和可靠地识别此类变化，以便我们能够赢得宝贵的时间采取适当的行动。在现有的大多数工作中，假设传感器可以采集多少样本以及何时采样没有限制。这一假设可能不适用于传感器网络中的许多应用，这些传感器由有限能量的电池或从环境中获得的可再生能源供电。重要的是设计新的具有能量约束的快速检测算法，以便所设计的算法能够在最小延迟的情况下利用电池或可再生能源供电的传感器来检测异常活动。能量限制对快速检测问题提出了巨大的挑战和独特的特征。设计自适应传感策略至关重要，这些策略依赖于从迄今采集的样本中提取的信息和电池的能量水平，以做出样本和检测决策。为此，利用最优停止理论的工具，该项目旨在实现以下目标：1)表征具有附加能量约束的问题的最优检测方案；2)了解与这些能量约束相关的性能损失；以及3)设计低复杂度但渐近最优的检测方案。为了实现这些目标，该项目将专注于两个研究推动力。在第一个研究推力中，该项目将专注于对传感器允许进行的观测总数有硬限制的情况。这一推力中设计的算法将对以有限能量的电池供电的传感器有用。在第二个研究推力中，该项目将专注于具有随机能量约束的场景。所设计的算法适用于可再生能源供电的传感器。

Wireless sensor networks are commonly deployed to monitor abnormal changes in their surrounding environment. These changes typically imply certain activities of severe consequences, such as structure failure or chemical/gas leak, etc. Quickest detection is a framework that focuses on the design of sequential detection algorithms to identify such changes as quickly and reliably as possible so that we can win valuable time to take proper actions. In most of existing works, it is assumed that there is no constraint on how many and when sensors can take samples. This assumption may not hold for many applications in sensor networks whose sensors are powered either by battery with limited energy or renewable energy harvested from the environment. It is important to design novel quickest detection algorithms with energy constraints so that the designed algorithms can be used to detect abnormal activities using sensors powered by battery or renewable energy with a minimal delay. The energy constraints present significant challenges and unique features to quickest detection problems. It is crucial to design adaptive sensing strategies that rely on information extracted from samples taken so far and energy level at the battery to make sample and detection decisions. Towards this end, using tools from optimal stopping theory, the project aims to achieve the following goals: 1) to characterize the optimal detection schemes for problems with additional energy constraints; 2) to understand the performance loss associated with these energy constraints; and 3) to design low-complexity but asymptotically optimal detection schemes. To achieve these goals, the project will focus on two research thrusts. In the first research thrust, the project will focus on scenarios with a hard constraint on the total number of observations that the sensor is allowed to take. The designed algorithms in this thrust will be useful for sensors powered by battery with limited energy. In the second research thrust, the project will focus on scenarios with a stochastic energy constraint. The designed algorithms are suitable for sensors powered by renewable energy.